Comments (1)
All these would be really useful, I think. You'd want:
- expected costs for a particular pipeline (either calculated directly, or via directing to some API or online costs calculator for the particular LLMs involved). In the case of multiple LLMs involved, this might get complicated i.e.
- expected tokens to be used for a particular pipeline, and split by LLM / task (so you can manually make the calculation yourself if you want)
- add a 'max_budget' option for a param, which would either stop when you hit that estimated cost, or it wouldn't run the pipeline and would suggest ways to downscale the amount of work your pipeline does in order to stay within the costs?
- expected_time is maybe interesting, but probably hard to do. But I agree it'd be pretty useful if you could get a ballpark figure.
You might even consider abstracting some of this out into some online calculator tool where you just get the number of tokens for a raw data file and then for a typical task you could get a rough estimate of both time and cost for typical scenarios (i.e. gpt3.5 or claude-haiku etc). Just from a marketing standpoint having one of these HF spaces that do that might be attractive.
I think this'd all be really beneficial for the tool. In the end, you can only make very loose guesses right now based on small subsection runs and then try to extrapolate out.
from distilabel.
Related Issues (20)
- [FEATURE] Add `O1AILLM`
- [FEATURE] Saving/Loading of Distiset with S3 bucket (or locally) HOT 4
- [FEATURE] Improve cache writing
- [FEATURE] Add `requirements` in `Pipeline`
- [CI] Install both `minimum` and `latest` dependencies in CI HOT 2
- [IMPROVEMENT] `PrometheusEval` support for multi-turn evaluation HOT 1
- [FEATURE] Implement "Improving Text Embeddings with LLMs"
- [BUG] 501 (Not Implemented) Response when trying to load HF dataset HOT 7
- [BUG] "An attempt has been made to start a new process before the current process has finished its bootstrapping phase." HOT 1
- [BUG] reponse_format - Unexpected keyword argument HOT 1
- [FEATURE] Make `LoadHubDataset` more general to read local files HOT 3
- [DOCS] Update cite information HOT 3
- [FEATURE] Add functionality to read/pass HF_TOKEN from cache/env var.
- [DOCS] Update document structure and format
- [DOCS] Update document phrasing and funnel
- [DOCS] Add examples for the components gallery
- [FEATURE] show raw input/response within tasks
- [FEATURE] trying to reload a already loaded `Step` results in an ambiguous error
- [BUG] `TogetherLLM` only works with json response format
- [BUG] `EvolQuality._apply_random_mutation` only uses last character of response
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from distilabel.